Confidence Scoring

Every openremap identify result and every openremap scan line includes a confidence assessment — a measure of how reliably the system detected and identified the ECU binary, based on detection quality, extracted identity fields, filename heuristics, and map-structure analysis.

What confidence measures: The score reflects identification quality — how many expected identity fields were successfully extracted and how rigorous the detection cascade was. It does not measure whether the binary content has been modified. A tuned file with all ident blocks intact will score just as highly as an unmodified one.

Tiers

Example output

HIGH — Bosch EDC17, all identifiers present

  ── Confidence ─────────────────────────────────────
  Tier   HIGH
  Signal  +  detection evidence (5 tags: MAGIC_MATCH, IDENT_BLOCK, SIZE_MATCH, FAMILY_STRING…)
  Signal  +  canonical SW version
  Signal  +  hardware number present (0261209352)
  Signal  +  ECU variant identified (EDC17C66)

HIGH — Delphi Multec S, all identifiers present

  ── Confidence ─────────────────────────────────────
  Tier   HIGH
  Signal  +  detection evidence (5 tags: SIZE_MATCH, BOOT_BLOCK, POINTER_TABLE, FAMILY_STRING…)
  Signal  +  canonical SW version
  Signal  +  hardware number present (09391237)

SUSPICIOUS — stop and check

  ── Confidence ─────────────────────────────────────
  Tier   SUSPICIOUS
  Signal  +  detection evidence (2 tags: SIZE_MATCH, EXCLUSION_CLEAR)
  Signal  -  SW ident absent — no match key produced
  Signal  -  tuning/modification keywords in filename
  ⚠  IDENT BLOCK MISSING
  ⚠  TUNING KEYWORDS IN FILENAME

Signals

Each signal line shows what contributed to the tier. A + prefix raised confidence; a - prefix lowered it.

Detection quality baseline

The extractor's can_handle() collects evidence tags as it checks the binary — MAGIC_MATCH, IDENT_BLOCK, SIZE_MATCH, and so on. Each tag is worth 2–4 points and the total is capped, so the baseline reflects how much the extractor could actually prove:

The signal reads detection evidence (N tags: …), listing at most the first four tags. An extractor that runs four or more structural checks reaches the cap.

Software version

Other identity fields

The variant bonus is skipped when the reported variant is just the family name.

Cross-checks against the binary

Two more bonuses come from checking the extracted identity against the binary itself:

The coherence pair needs a cross-check report that only openremap analyze builds, so identify, scan and health never add or subtract these points. A third coherence case — a stale checksum on a tuned file — changes no points and only raises ⚠ CHECKSUM STALE (tuned file).

Filename signals

These are the only signals that relate to file originality rather than identification quality. They flag filenames that suggest the binary may have been modified or provide no identifying context.

Only one of the two applies: the tuning-keyword check runs first and, on a match, the generic-filename check is skipped.

Warnings

Warnings flag specific concerns, independent of the tier score:

How the score is calculated

Each signal carries a positive or negative delta. The deltas are summed into a raw numeric score, which is then mapped to a tier:

The raw score is available in JSON output (openremap identify --json) under confidence.score, alongside the tier string and the full signals array — each entry carrying its own delta and label. The terminal prints the tier upper-case (HIGH); the JSON value is title-case ("High").

Family field profiles

An ECU family can declare a field profile — the set of identity fields that the ECU architecturally stores in its binary. Fields not in the profile are never penalized when absent.

For example, the IAW 1AP family only stores a calibration fingerprint. It does not contain a discrete software version or hardware number in the binary. Under the old scoring system this absence would have incurred penalties, dragging the tier down unfairly. With field profiles, the absence of SW and HW for IAW 1AP scores 0 instead of a penalty, allowing these families to reach their natural tier based on the fields they do provide.

Families like Bosch EDC17 declare SW, HW, variant and calibration, so all four are expected and scored accordingly. Bosch ME7 declares SW, HW and calibration but not variant — a missing variant is never held against it. Siemens EMS2000 declares an empty profile, so no field is ever expected from it and none of the absence penalties apply.

A family with no profile registered is treated conservatively: every field is assumed to be expected, so an absent SW is flagged rather than silently ignored.

Detection quality

The extractor's can_handle() varies in rigour across families. One that checks magic bytes, block boundaries, address maps and internal checksums collects many evidence tags; a shallow heuristic that only compares the file size and one byte pattern collects few.

That difference is what the baseline measures — a tag sum, not a label:

• Many structural tags (a deep, multi-check cascade such as Multec S or IAW 1AV): reaches the +20 cap. A match is very likely correct.
• Several mixed tags (a family that checks size, a marker string and an ident block): lands in the +10–16 range. Solid but not exhaustive.
• One or two soft tags (a size match plus an exclusion check): +4 or less. Plausible, less certain.

The Detection quality baseline above has the per-tag weights, and its note explains the deprecated static detection_strength.

Manufacturer-aware scoring

The confidence system is manufacturer-aware — each manufacturer has its own canonical software version pattern, and each ECU family declares its own field profile. This ensures that all manufacturers can reach HIGH tier when their binaries contain the expected identifiers.

Canonical SW version formats by manufacturer

Any SW version that matches its manufacturer's canonical format earns +30. A SW version that is present but does not match the expected format earns +15. This replaces the previous approach where only Bosch 1037-prefixed versions received the full canonical bonus and all other manufacturers were capped at the non-canonical score.

Using confidence in practice

# Full per-signal breakdown for a single file
openremap identify ecu.bin

# Confidence as JSON — score, tier, signals array, warnings
openremap identify ecu.bin --json

# Triage an entire folder — confidence tag on every line
openremap scan ./my_bins/

# Export confidence scores and warnings for every file to CSV
openremap scan ./my_bins/ --report report.csv

A SUSPICIOUS result is not a verdict — it is a prompt to look closer. Run openremap identify on any flagged file for the full signal breakdown before deciding whether to use it.

← Back to CLI reference